OPTIMIS – TOWARDS HOLISTIC CLOUD MANAGEMENT Johan Tordsson, Department of Computing Science & HPC2N, Umeå Universitet
2 OPTIMIS: BACKGROUND AND MOTIVATION What? IP, Call 5, 10.4 M€ budget, 13 partners (8 academic) Why? Multiple cloud models, definitions, etc. Our view: Private clouds are common practice within the next few years Additional resources to handle load peaks etc. are provided by public cloud(s) No one-size-fits-all solution to cloud provisioning Need for common abstractions, tools, and methods for various scenarios
Roles & Challenges New challenges –New customers –New business models –New collaboration forms –New requirements
4 F IVE CONCERNS FOR FUTURE CLOUDS 1. Dependable sociability Management based on non-functional aspects Foundation for eco-system of providers and consumers of cloud services 2. Many cloud architectures Private, bursted, federated clouds, etc. 3. Service life cycle management optimization Construction, deployment, operation 4. Adaptive self-preservation Self-* management with respect to functional and non- functional aspects 5. Market and legal issues Identify business opportunities and legislative concerns
5 1. D EPENDABLE SOCIABILITY Beyond cost-performance tradeoffs Tools for measuring & prediction of TREC: Trust Reputation-assessment of actors (SPs, IPs, etc.) Transitivity aspects Risk Probability of something (bad) happening … … and the consequences Identification, assessment, monitoring, treatment Eco-efficiency Monitor and predict power, PUE, CO2, etc. Compliance to standards and legislations Cost Need for economical models beyond list prices Required to balance the above 3 factors
Multi-clouds Federated Clouds Infrastructure Provider Bursted Private Clouds 2. M ULTI -C LOUDS : T HREE B ASIC S CENARIOS Infrastructure Provider Service Provider Broker Infrastructure Provider Service Provider Infrastructure Provider Private infra- structure
Cloud providers Eco-System – Programming Model – Services Composition [Legacy & New] Construction – Self-management – Risk Evaluation – Eco-efficiency –Data Management Internal Cloud Operation Optimization Plus: – Multi-clouds – Federated clouds – License Management – Eco-efficiency Evaluation – Security External Cloud Operation Optimization – Risk Assessment – Trust Circle – Eco-efficiency Evaluation – Economic factor Deployment Optimization 3. S ERVICE LIFE CYCLE
8 4. A DAPTIVE S ELF - PRESERVATION Clouds are complex and environments change rapidly We need Automatic self-* management of infrastructure self-configuration self-healing self-optimization Holistic view Cannot do management of services, VMs, data, etc. in isolation Self-management based also on non-functional aspects Trust, risk, eco-efficiency, and cost Policy-driven management Adaptable and replacable policies
5. M ARKET AND L EGAL ISSUES Cloud eco-system new and currently evolving Opportunities for new roles, business models, relationships, value chains, etc. Legal concerns Acquisition, location, and transfer of data Across borders and legal domains Data protection and security mechanisms needed (CS) Research problem How to design mechanisms to be used to implement currently not known policies?
O UR APPROACH – THE OPTIMIS T OOLKIT Addresses the five challenges Generic toolset to support multiple cloud architectures Reusable and configurable components Incorporates TREC-management and self-* abilities Supports full service life cycle Data protection capabilities
OPTIMIS S YSTEM MODEL What is a service (in OPTIMIS)? Any functionality offered to clients over a network Delivered through one or more VMs Elastic #VMs change dynamically during operation Defined by SP in a service manifest VM images (OVF) SLAs w.r.t. elasticity (service-specific KPIs) Tresholds with acceptable levels of trust, risk, eco- efficiency, and cost Deployed by SP in IP(s) Operated by IP(s)
OPTIMIS T OOLKIT OVERVIEW Four main groups of components Basic Toolkit SP tools IP tools Tools usable by both SPs and IPs
B ASIC T OOLKIT Monitoring Core functionality for self-managed systems 3 levels Services Virtual infrastructure Physical infrastructure Tools for measurement and prediction of Trust Risk Cost Eco-efficiency Security Identity management, etc. to handle interconnection of clouds
SP T OOLS Programming model Implement new service components Integrate existing ones IDE + runtime for workflow style applications License management Integrate license-protected software in services Challenges: Elastic services Migrating services
SP T OOLS ( CONT.) Service Optimizer (SO) Overall management of services Tracking state and deployment(s) Performance monitoring Re-deployment Contextualization mechanisms Dynamic runtime setup of VMs and services, with respect to networking etc. Two step process: Preparation Attach boot-scripts in ISO image and couple this with VM image Self-contextualization Booting from ISO-image
IP T OOLS Admission Control (AC) Capacity planning and safe overbooking Accept incoming service request or not? + Increased revenue - Added provisioning costs ? Implications for already hosted services Services are elastic Degree of elasticity differ Time and duration of spikes differ Similar problems Network bandwidth multiplexing Selling airline seats Long-term capacity planning (cf. scheduling)
IP T OOLS (C ONT.) VM Management VM lifecycle management Scheduling: optimal mapping of VMs to physical hosts in an IP across multiple clouds Federation and bursting When? Admission of new service, upon elasticity, faults, periodically Optimal? SP perspective: Performance (hosts, VMs), cost, guarantees, TREC, etc. IP perspective: Provisioning cost, consolidation, isolation, SLA violations, etc.
IP T OOLS ( CONT.) Fault Tolerance Engine Automatic VM checkpointing and restart Intervals configurable Cloud Optimizer (CO) Combines monitoring and prediction with IP-level engines to perform self-management Overall decisions related to local vs. bursted/federated VM placement etc. Policy reconfiguration
C OMMON TOOLS FOR SP S AND IP S Service Deployment Optimizer (SDO) Coordinates service deployment process Discovers and filters IPs, negotiates SLAs, assesses TREC-factors, contextualizes services, uploads data, deploys services Service deployment (SP to IP) Private cloud + multi-cloud service deployment VM placement (IP to IP) Cloud bursting + federation Data Management Transfer of VM images and service data for deployment; SP to IP, and IP to IP Manages distributed file system for service applications Automatic re-location of service data across federated IPs
C OMMON T OOLS ( CONT.) SLA Management (CloudQoS) Creation and monitoring of SLAs WS-Agreement term extensions for TREC Negotiation primitives (WS Agreement Negotiation) Elasticity Engine Feedback controller for automatic and proactive VM allocation to meet peaks and lows in demand More about this one later…
E XAMPLE : S ERVICE DEPLOYMENT (SP SIDE )
E XAMPLE : S ERVICE DEPLOYMENT (IP SIDE )
E XAMPLE : SERVICE OPERATION (IP SIDE )
OPTIMIS T OOLKIT DEPLOYMENT ILLUSTRATIONS Bursted private clouds IP SP SDO SO CO ACSDO SO Private Cloud CO AC IP CO ACSDO SO IP SP SDOSO Federated clouds CO AC IP CO AC IP SP SDO SO Multi-clouds CO AC IP CO AC IP
F URTHER F UTURE D IRECTIONS Use cases: Programming model Service construction/composition Examples in ERP/CRM (SAP) and bio-informatics Cloud bursting Outsourcing based on TREC Interoperation with OPTIMIS and non-OPTIMIS Ips E-Education test cases Cloud brokering, a broker: Acts as IP to SPs Acts as SP to IPs Is independent? Provides value-added services? Infrastructure Provider Service Provider Broker
C URRENT & F UTURE DIRECTIONS ( CONT.) OPTIMIS (the project) : June 2010 … May 2013 Basic plumbing in place Algorithmical improvements next focus TREC-aware self-* management policies Holistic management, BLO-driven IP- and SP- operation Experimentation needed Open for collaborations
A CKNOWLEDGMENTS : THE OPTIMIS CONSORTIUM